Related link: http://linuxworldexpo.com/linuxworldny03/V40/index.cvn
The environment of a trade show is blaringly inorganic. But I sometimes felt at LinuxWorld as if I was running back and forth through a garden, checking the progress of each sapling. I used the three brief days of the show to track the growth of Linux in several types of applications:
Proponents of embedded Linux seem satisfied with its rate of adoption. There’s no way to be sure how many companies are using it, given the lack of licensing and the tendency of companies to treat their choice of embedded system as a competitive secret. But I heard lots of praise at the conference for the qualities of Linux that help it win over other embedded systems.
Chief among the qualities cited are Linux’s sophisticated network stack, which supports an impressive range of the latest standards from Bluetooth to IPv6. I heard, at a presentation on Carrier Grade Linux by Glenn Sieler of
MontaVista Software, that Linux’s support for IPv6 was attracting the interest of the telecom industry (which he estimated as 30% of the embedded Linux market), particularly in Korea and Japan where they cannot ignore the looming threat of a shortage of IPv4 addresses.
Sieler laid out a laundry list of Linux’s advantages for telecom devices, both technical and social:
- Ample technical support.
- A wide range of supported hardware, offering easy portability of applications to new platforms.
- Availability of source code.
- Royalty-free use.
- Powerful development tools.
- Small footprint and modular design.
One key player in embedded Linux is the uClinux project (where the “u” stands for “micro”), a port of Linux to chips without memory management units. Building a microprocessor without an MMU, according to uClinux maintainer Greg Ungerer, allows it to be cheaper and even a bit faster than conventional chips. The disadvantages of doing without an MMU, such as static-sized stacks and the lack of memory protection, are often unimportant to embedded developers. uClinux will be integrated into the core Linux 2.6 distribution.
Ungerer and another uClinux maintainer, David McCullough, work at
a company that makes a feature-rich firewall and routing appliance so compact you can practically twirl it on one finger.
Sangoma goes even farther in shrinking the router, providing a single network interface card that, according to founder David Mandelstam, allows typical desktop systems to replace specialized routers and telecommunications equipment. Mandelstam, one of the more passionately evangelical CEOs I met, declared that he wanted to demystify routers. Vanilla Linux already provides routing and firewall software as full-featured as most people would want, and typically runs with much more memory and CPU power than high-priced, specialized routers. As for protocols, Sangoma cards support ADSL, Frame Relay, ATM–you name it. And while it plugs into many types of systems, Mandelstam considers Linux the best choice for this application.
The consensus among most observers, outside of the most enthusiastic Linux proponents, is that it will take a long time for Linux to be adopted by the average knowledge worker or home user. Thus, companies seem to be focusing on selling into the enterprise, with desktop Linux as an option to provide a consistent environment with Linux on servers. I believe this strategy lay behind Red Hat’s recent announcement that it would no longer sell its personal workstation version in shrinkwrapped CD form. SuSE is also focusing on the enterprise. However, see the successes I cited for the
Linux Terminal Server Project
in Thursday’s weblog.
While most uses of Linux examined in this article consist of entries into existing markets and applications, clustering is an area where it has opened up exciting new possibilities. Although clustering existed before Linux, it has broken new ground by enabling the use of cheap commodity hardware without software license fees.
It appeared to me at LinuxWorld that Linux clustering has become almost ubiquitous where solutions for large enterprises are found. IBM advertises that its DB2 database management system (the database behind
can run on a cluster of shared-nothing (independent storage) Linux systems. Stalker’s CommuniGate (mentioned in
Thursday’s weblog) uses clusters. Many more instances could be cited.
Emic Networks, which markets Linux clustering solutions for Apache and MySQL to permit load-balancing and high availability, imposes very stringent requirements on its clusters. Each node within the cluster can handle a MySQL update and inform the user that it has finished, but before the cluster can handle any further queries, the update is propagated to all the other nodes so they move forward in lockstep. In this manner, different parts of a transaction (as well as overlapping queries and updates) can be handled by different nodes without the danger of dirty or phantom reads. A reliable protocol is layered on top of a UDP broadcast to accomplish the synchronization.
Incidentally, the appearance on Monday of an article in the New York Times promoting the importance of traditional Cray-style supercomputers has no bearing on this Linux clustering trend. It’s not surprising that certain applications traditionally run on big-metal supercomputers require that architecture. The important point is that commodity hardware and license-free software is discovering an unexpectedly wide range of new applications.
Surprised? I wasn’t expecting originally to include a section on mainframes either. But Linux, which has always been a way to extend the life of old Windows and Macintosh hardware, has been thriving on legacy IBM mainframes for the same reason. According to VP of Marketing for Linuxcare, Beat Knecht, sites are even buying new mainframes to run Linux. IBM continues to push its mainframe and even announced a port of Lotus’s Domino server to Linux on the zSeries mainframe.
While Linux has been ported to the bare hardware, it is apparently more common to run several instances of Linux on top of IBM’s native MVS. Knecht explained that Linux, even though it is a multiuser operating system, does better running only one service at a time when the large applications that are popular nowadays are involved, such as databases and web servers.
Linuxcare, one of the earliest commercial names in the Linux space, has been extending the life of mainframes and doing a pretty good job of extending its own life as well by turning into a vendor of management software called Levanta for Linux on the mainframe. What was remarkable to me is that some 45 conference attendees took the trouble to cross the street and slip into the gothic industrial-nostalgia interior design of the W Hotel to hear CEO Avery Lyford’s pitch.
LinuxWorld is an important show, a good place to pick up what’s hot fast. It’s also loud, high-pressure, and sometimes vulgar. The temptation to hype one’s product may be overwhelming, just to make it stand out among the din of competing messages, and I may have succumbed once or twice to the heady brew.
Among the marketing materials that IBM handed out at LinuxWorld was the factlet that the world produced two exabytes of information in 1999. I hope that the weblogs I’ve generated at this conference add a little to this information. The other weblogs I wrote at the conference are:
What other trends in Linux and free software should be noted?